Self-perception theory is an account of attitude change developed by psychologist Daryl Bem. It asserts that we only have that knowledge of our own behavior and its causation that another person can have, and that we therefore develop our attitudes by observing our own behavior and concluding what attitudes must have caused them.

Self-perception theory differs from cognitive dissonance theory in that it does not hold that people experience a "negative drive state" called "dissonance" which they seek to relieve. Instead, people simply infer their attitudes from their own behavior in the same way that an outside observer might. In this way it combines dissonance theory with attribution theory.

Bem ran his own version of Festinger and Carlsmith's famous cognitive dissonance experiment. Subjects listened to a tape of a man enthusiastically describing a tedious peg-turning task. Some subjects were told that the man had been paid $20 for his testimonial and another group was told that he was paid $1. Those in the latter condition thought that the man must have enjoyed the task more than those in the $20 condition. Bem argued that the subjects did not judge the man's attitude in terms of cognitive dissonance phenomena, and that therefore any attitude change the man might have had in that situation was the result of the subject's own self-perception.

Also, cognitive dissonance theory cannot explain attitude change that occurs when there is no upsetting dissonance state, such as that which occurred to subjects in studies of the overjustification effect.

Whether cognitive dissonance or self-perception is a more useful theory is a topic of considerable controversy and a large body of literature, with no clear winner. There are some circumstances where either theory is preferred, but it is traditional to use the terminology of cognitive dissonance theory by default.